What is the feature?
I have been using yolov8 pose model from Ultralytics and noticed that there is a noice or jitter in the spatial position of the keypoints predictions. It varies from 20 to 50 pixel radius each time it gives predictions in temporal frames with same object. Wondering if there is any approach or model that avoids this kind of noice and gives predictions almost same spatial position?
Any other context?
No response